170 likes | 292 Views
rostudel.com. ROSTUDEL. OPTIMIZATION AS A SERVICE. rostudel.com. OAAS (Optimization As A Service). OR : Theory + code (libraries+modelers) Architecture Consulting R&D : Seminars, University. rostudel.com. OR CRISIS ?. 4 success stories : we can actually use OR for realistic problems
E N D
rostudel.com ROSTUDEL OPTIMIZATION AS A SERVICE
rostudel.com OAAS (Optimization As A Service) • OR : Theory + code (libraries+modelers) • Architecture • Consulting • R&D : Seminars, University
rostudel.com OR CRISIS ? • 4 success stories : we can actually use OR for realistic problems • OAAS can brand OR to Operational Resources, not only to a dedicated PHd Team
rostudel.com 1st Story : Scheduling People • Retail organizations : employees, activities, constraints, workload, daily patterns, week scope => timetable at a 15 minutes detail • LP to cover workload + BALAS to linearize disjunctions • Dynamic Programming + CP + dominacy rules to generate Shift • Branch and Price when final MIP isn’t enough
rostudel.com Timetable
rostudel.com 2nd story : Smooth Cars Purchase with split batches. Balance holding cost and day demand JAN JUN
10 12 21 rostudel.com Demande constante de 96 véhicules par jour couverte par deux batchs: une arrivée de 43 véhicules un départ de 12 véhiculesune partie fixe de 63 véhicules
rostudel.com Sometimes OR is painless ! • Solves to optimality with a MIP • Easy to compare to previous solution • Easy to sell • Easy to extend => more batches, wider time window, plug with long term purchase application
rostudel.com 3rd Story : DIGITAL PRODUCTION • A job shop problem with machines, stockers, people and trolleys • Semiconductors : Billions € spent in the plant, almost no tool for optimization before “FAB300”
Scenario :déroulement processus (1 lot, 1 niveau) LE2 LE1 TM7 11 TM7 14 TM7 12 TM7 13 LE3 ST108 ST107 TM7 15 TM7 01 TM7 03 TM7 04 TM7 09 TM7 07 TM7 08 TM7 05 TM7 10 TM7 02 TM7 06 CD5 OV1 CD1 ST114 PRI_RS203 CD6 OV2 CD2 CD7 CD3 OV3 ST? CD8 OV4 CD4
rostudel.com NP is not good for CRM • CP dedicated ILOG SCHEDULER • LNS • Problem tackles only small instances • Improve LNS • Use customer heuristic • Forward : “warm rescheduling” with “encours”
rostudel.com 4th story : Generating Cab Moves Covering with Shifts TRIPS MAY REQUIRE A SPECIFIC SET OF SKILLS
rostudel.com VRO-1 : Duties LEE-YOR 0D242 LON-YOR 0D224 Base 1 Base 1 YOR-BIR 0D289 Base 2 Base 2
OPL CP syntax to get rid of cycles nbLegs[Source] == 0; forall(c in CaBMovesSource : c!=Source) { (prev[c]!=c) => (nbLegs[c] == 1 + nbLegs[prev[c]]); (prev[c] == c) => (nbLegs[c] == 0); }
rostudel.com Use OPL5.2 : one week compare to 3 years ! Solver Component Simplex (LP) (Ilog Cplex) MIP (Ilog Cplex) Constraint Programming (CP) (Ilog Solver) Specific constraints Column Generation « Shortest path » Algorithm
OAAS other stories… • Work together : one consulting company XXXX • Military Communication • QOS expected : assign radio gears under constraints • Perfect MIP model (100,000 binary variables, good solution after 30 sec) • Nobody to code : J2EE expected with DB • XXXX ROSTUDEL TELASIDE • SPRING + HIBERNATE (MySQL...) + CPLEX JAVA
MOVING FORWARD • “People on the Operations” enjoy mature code and products • A new market for consulting and support , besides traditional OR consumers • The fish is ready to byte : the door is open for “Serious OR” with “Huge” volume of data. • TV Add Campaign MIP : • 100,000 rows • 155,000 variables (100,000 booleans) • 2 millions Non Zeros rostudel.com Questions ?